The setting between allowed and removed that Meta admits exists

Your post is still up. You can see it. Your followers can see it. But it stopped showing up in the hashtag page ten minutes after you posted it. You check again an hour later. Still nothing. You search for it by caption text. Not there either.

So you search for why. An answer is waiting. You have been shadowbanned. Instagram is hiding your posts from everyone who does not already follow you. The term is everywhere. It is in thousands of posts and videos. Instagram's head says it is not real.

He is both right and wrong.

We went looking for what Meta publishes about this. Not what creators say. What the company itself says. In its own documents. Using its own words. Meta does not call it shadowbanning. It calls it something else. And it describes it in detail. Instagram also publishes a status that tells you whether your content is eligible to be recommended. That status covers one part of what follows. It does not cover the rest.

Here is what we read. Meta publishes pages explaining how it handles content. These sit on its Transparency Center website. That is where the company posts its policies. We read the pages about reducing distribution. We read Instagram's creator blog. We read the Community Standards. We also read what Adam Mosseri said about shadowbanning in 2020. He runs Instagram. And we read an investigation by The Markup. They are a nonprofit newsroom. They ran experiments on Instagram in early 2024. They found content being hidden from hashtag pages. Users got no notice.

Meta has a framework. It is called Remove, Reduce, Inform. It describes three things the company does to content. Remove means take it down. Inform means add a label or context. Reduce means something in between. Content that does not break the rules but still gets shown to fewer people. That middle category exists. Meta admits it. The company has published that framework since 2016.

What follows is a reading of Meta's own documents. It covers Instagram. We do not speculate. We quote what Meta wrote. We report what The Markup found. Every source below is linked, dated and quoted.

Meta publishes a three-part framework: Remove, Reduce, Inform

Start with the framework itself. Meta announced it publicly in April 2019. The announcement ran on the company's own blog. The headline was "Remove, Reduce, Inform: New Steps to Manage Problematic Content". But the page says the company has used this strategy since 2016. It just had not named it publicly yet.

The three parts work like this. In Meta's words from that 2019 post.

Remove. Content that violates Community Standards gets taken down. Community Standards are Meta's rules for what you cannot post. The rules cover hate speech, bullying, and other violations. Meta uses both automated systems and human reviewers to find it.

Inform. Meta adds context to some content. So people can decide for themselves. This includes fact-check labels. And background on who published something.

Reduce. This is the one that matters here. Meta's 2019 announcement describes it in one sentence:

while we allow it to be posted on Facebook, we want to make sure it's not broadly distributed. — Meta, Remove, Reduce, Inform: New Steps to Manage Problematic Content, April 2019

Content in this category stays up. Your followers can still see it. But the platform reduces how far it travels. Meta calls this content "problematic". The announcement gives examples. Misinformation rated false by fact-checkers. Clickbait. Clickbait means headlines that promise more than the article delivers. Later in the same post, Meta describes reducing distribution of Groups. The ones that repeatedly share false information.

That was 2019. The framework still governs today. Meta's current Transparency Center page about lowering distribution opens with the same three-part structure. The page is titled "Reducing the distribution of problematic content". Retrieved 17 August 2026. It describes content that "doesn't violate the Community Standards, but might still be problematic or otherwise low-quality". For that content, Meta writes: "Meta may reduce its distribution."

No shadowban is named. The behavior is described anyway. Content that stays up but travels less far.

Borderline content: Meta's term for the middle category

Meta has a name for some of this content. The company calls it borderline. That term appears on a Transparency Center page. The page is titled "Types of content that we demote". Retrieved 17 August 2026. The page lists categories of content whose distribution Meta reduces.

The page names borderline content. It gives four examples. All quoted from the page:

• "photos of people posing in a sexually suggestive manner"
• "speech that includes profanity"
• "borderline hate speech"
• "gory images"

The page describes how Meta handles it:

We make this content less visible for those who prefer not to see it. — Meta Transparency Center, Types of content that we demote

Notice what that sentence does not say. It does not say the content violates a rule. It says some people prefer not to see it. And for those people, Meta makes it less visible. That is the mechanic. Content is sorted into a category. The category determines how far it travels.

The same page says how much the distribution gets reduced. It varies. Three things control it. How many times the poster has violated policies before. How confident Meta's systems are about the classification. And other factors the page does not name.

One more detail from that page. Enforcement is "generally global". But Meta may adjust it "temporarily" in a specific region. Or during a critical event. That sentence matters for the investigation described later in this post.

Instagram uses two separate rule sets

Instagram splits the rules in two. One set governs what gets removed. The other governs what gets recommended. The distinction showed up in a blog post Instagram published for creators. The date was 8 June 2023. The post is titled "Recommendations on Instagram: What creators need to know".

The two rule sets are:

Community Guidelines. These apply to all content on Instagram. Content that violates them gets removed entirely.

Recommendation Guidelines. These control what Instagram shows to people who do not follow you. Content that violates these guidelines can still be seen by your followers. But it will not be recommended to non-followers. Recommended means shown in Explore, in hashtag pages, or in the Reels tab.

The post states the distinction in plain words:

Content that goes against our Recommendation Guidelines can still be distributed normally to your followers, but we try not to recommend it to non-followers. — Instagram for Creators, Recommendations on Instagram: What creators need to know, 8 June 2023

Two different systems. Two different outcomes. A post that breaks Recommendation Guidelines stays up for followers. But it stops traveling to new people. A post that breaks Community Guidelines gets removed for everyone.

The post gives examples of what violates Recommendation Guidelines but not Community Guidelines:

• Violence depictions, such as fights between people
• Promotion of regulated products like tobacco or vaping products
• Sexually suggestive content, including images of people in sheer clothing

These things do not get you banned. They get you filtered out of recommendations. Your existing followers still see the post in their feeds. New people do not see it in Explore. They do not see it in hashtag pages. They do not see it in the Reels tab.

The post also says violations can affect your whole account. Not just one post. If you frequently post content that violates Recommendation Guidelines, your entire account may become ineligible for recommendations. Same if your bio or profile picture violates them. Not just the violating posts. Everything you post.

That is the mechanism everyone calls shadowbanning. Meta does not use that word. It describes the same outcome under a different name.

What lands content in the reduced category

Meta publishes lists of what gets reduced distribution. The Transparency Center page on lowering distribution names three principles. Each one covers different kinds of content.

Responding to people's direct feedback. Meta reduces distribution of content that users signal they do not want to see. This includes:

• Clickbait links
• Engagement bait. That means posts that explicitly ask for likes, shares, or comments.
• Links to websites that are slow to load, ad-heavy, or broken
• Low-quality comments that are copy-pasted repeatedly

Incentivizing high-quality content. Meta reduces distribution of:

• Content with limited originality that repurposes material from other sources
• Low-quality videos that abuse video or live formats
• Misinformation rated false or altered by third-party fact-checkers

Fostering a safer community. Meta reduces distribution of:

• Content from creators who repeatedly violate policies

One category sits outside those three. Meta also reduces distribution of "Content Likely Violating Community Standards". That is content Meta's systems flag as probably breaking a rule. But the systems are not confident enough to remove it outright. So it stays up, with reduced reach. Until a human reviewer can check it.

All of this comes from the same Transparency Center page about lowering distribution. Retrieved 17 August 2026. Every category named above is something Meta admits reducing. Not removing. Reducing.

What "reduce" means, and what Meta will not say

Here is what Meta publishes about how much distribution gets reduced. Almost nothing.

The pages describe the behavior. Content "may" get reduced distribution. The degree varies based on past violations. And on system confidence. But no page we read states a number. Not a percentage. Not a multiplier. Not a floor or ceiling. Just that it happens.

The Transparency Center page uses one phrase repeatedly. "Reduce its distribution". That phrase appears on multiple pages. It never comes with a measurement.

So we know content can be placed in a category. That category reduces how far it travels. We do not know by how much. We do not know for how long. We do not know what signals move content into that category. Or out of it. Meta publishes the framework. It does not publish the weights.

One detail does appear. From the types-of-content-we-demote page. The degree of reduction depends on the number of previous violations. So repeated violations compound. A first offense reduces reach by some amount. A second offense reduces it more. The page states that. It gives no numbers for either.

The Markup found systematic suppression of content

Meta describes reducing distribution in theory. The Markup tested what happens in practice. They are a nonprofit newsroom. In February 2024 they published an investigation. The title was "Demoted, Deleted, and Denied: There's More Than Just Shadowbanning on Instagram". They spent four months testing Instagram's content moderation. They created 70 test accounts. They posted more than 200 times. They invented two new hashtags. So they could watch exactly what appeared and what did not.

The finding that matters most. Nongraphic photos depicting war were hidden from hashtag pages. At a much higher rate than other content. The Markup posted images of destroyed buildings, soldiers, and military tanks. None showed casualties or blood. None violated Community Guidelines. They posted these to a custom hashtag called #markuptest.

The results:

• Only 9 of 29 war photos ever appeared on the #markuptest hashtag page
• 158 of 172 non-war posts from the same accounts appeared on the same hashtag page
• War imagery was 8.5 times more likely to be hidden than other content

The Markup received no notification that these posts were being demoted. The posts stayed up. Followers could see them. They just did not appear in the hashtag page. That is reduced distribution. Documented and measured.

The investigation found more. Instagram deleted entire captions containing #markuptest. Without telling users. The photo posted successfully. The text disappeared. No notice. The Markup says this happened "dozens of times".

The #markuptest hashtag itself became unreliable. Sometimes it showed more than half of the posts made to it. Other times it showed less than a third. For some accounts, the hashtag search displayed "100+ posts". But clicking through showed "No results found".

Instagram also hid comments. Without telling the commenter. The Markup posted ten comments designed to look like spam. Eight of those comments were visible to the person who posted them. But invisible to everyone else. The commenter saw their own comment. No one else did. Not even in "View hidden comments". The commenter received no notice that this had happened.

Meta spokesperson Dani Lever responded to The Markup's findings. She said: "any implication that we deliberately and systemically suppress a particular voice is false." She did not address why war photos were less likely to appear on hashtag pages. She did not address why captions were deleted without notice.

One detail from the investigation is worth holding on to. The Markup found no evidence that pro-Israel or pro-Palestine accounts were treated differently from each other. Both sides saw similar visibility for ordinary content. The suppression was about the subject matter. Not the ideology. War imagery specifically was hidden at a higher rate.

The Markup states their limits clearly. All test accounts were brand new. They were accessed from the same IP addresses. The sample size was small. It did not allow for statistical significance testing. Instagram's policies changed multiple times during the research period. None of that changes the core finding. Content was systematically hidden from discovery. And users were not told.

Mosseri says shadowbanning does not exist

Adam Mosseri runs Instagram. In 2020 he said shadowbanning is not real. His exact words, from an Instagram Q&A:

Shadowbanning is not a thing. If someone follows you on Instagram, your photos and videos can show up in their feed. — Adam Mosseri, Instagram Q&A, 2020

That statement is true in the way he said it. If someone already follows you, your posts can show up in their feed. Shadowbanning, as most people understand it, means your followers stop seeing your posts. Mosseri says that does not happen.

What he does not say: posts can stop reaching non-followers. That is a different claim. And it is exactly what Meta's Recommendation Guidelines describe. Content that violates those guidelines "can still be distributed normally to your followers, but we try not to recommend it to non-followers." That is Instagram's own creator blog. From 2023. Quoted earlier in this post.

So Mosseri is right. Instagram does not hide posts from your existing followers. Meta is also right. Instagram reduces distribution to non-followers. For content that violates Recommendation Guidelines. Both things are true. The word "shadowban" does not appear in Meta's documents. The behavior does.

Meta's Oversight Board noticed this too. The Oversight Board reviews Meta's content decisions. It is funded by Meta. But it operates independently. In June 2026 it examined how Meta penalises accounts. Engadget's report of those findings describes what Instagram does instead of the account restrictions Facebook uses:

Instead, Meta restricts accounts from Instagram's livestreaming feature or will remove their account from recommendations (which Instagram users often refer to as a 'shadowban'). — Engadget, Even Meta's Oversight Board thinks its rules for banning accounts are baffling, 4 June 2026

The parenthetical is Engadget's, not the Board's — a reporter's gloss for readers, not a term the Board adopted. What the reporting establishes is the practice itself. Meta removes accounts from recommendations. That is what reduced distribution means at the account level.

So here is the shape of it. Shadowbanning is not an Instagram feature. Instagram does reduce distribution. Of content and accounts that do not violate Community Guidelines. But do violate Recommendation Guidelines. Or get classified as borderline. That reduction is invisible to the user. No notice. No flag. No explanation. It matches what most people describe when they say they have been shadowbanned. Meta just does not call it that.

What The Markup's findings mean for the engagement theory

The common theory of Instagram reach is simple. Engagement signals to Instagram that a post is good. So Instagram shows it to more people. Get enough of it early and the post compounds.

That theory assumes Instagram is sorting posts by engagement. But The Markup's investigation found content being hidden based on its classification. Not its engagement. War imagery was suppressed even when it came from accounts with followers. Even when it had captions and hashtags that other posts used successfully. The suppression happened at the content level. Before engagement could accumulate. Before reach could compound.

If Instagram classifies your content as borderline, engagement may not matter. Or as violating Recommendation Guidelines. Or as belonging to one of the categories Meta says it reduces. The reduction happens upstream of that. At the point where Instagram decides whether to surface the content at all.

We do not know that for certain. Meta does not publish how these systems interact. We do not know if strong engagement can override a borderline classification. We do not know whether anything a creator does after posting changes a classification already applied. What we do know is this. The Markup documented content being systematically hidden. Regardless of what users did with it.

What we could not determine

Five things. Named so no one thinks we checked and stayed quiet.

How much distribution actually gets reduced. Meta says it reduces distribution. It never says by what percentage. Not on any page we read. Not for any category. We report that the reduction happens. We cannot report its size.

How long the reduction lasts. No page we read states a duration. Whether borderline content stays borderline forever. Or gets reassessed. Or ages out of the classification. Nothing published.

What moves content into borderline status and what moves it out. Meta describes the categories. It does not describe the signals the classifier uses. We do not know what threshold a post crosses to become borderline. We do not know what would remove that classification.

Whether engagement can override a borderline classification. Meta's ranking cards say engagement is a signal. Meta's distribution guidelines say borderline content gets reduced distribution. We do not know which one wins. Or whether they interact at all. The Markup's findings suggest classification dominates. That is evidence. Not proof.

Whether Recommendation Guidelines violations are permanent or temporary at the account level. Instagram's creator blog says repeated violations can make an entire account ineligible for recommendations. It does not say for how long. It does not say whether fixing the violations restores eligibility. It does not say whether deleting old posts helps. We found no page that answers those questions.

How we did this

Scope: this post reads published documents and one investigation. No data collection of our own. No experiment.

How we read them: every source was opened directly on 17 August 2026. Every quoted sentence was copied off the page as a reader sees it. Seven of the sources are Meta's own pages. The Transparency Center pages and the Instagram for Creators blog posts. One is Meta's 2019 announcement on the company blog. Two are news articles reporting on primary sources. One is The Markup's investigation. Which includes full methodology. We could not access one academic paper on shadowbanning behind a paywall. Its findings appear only from a publicly accessible summary.

Limits: eight of ten sources are first-party statements. All from Meta. A Transparency Center page is a description Meta chose to publish. Not an audited account. Meta rewrites these pages without saying when. So every quotation here carries the page's date stamp from the day we read it. The Markup's investigation was conducted over four months. In late 2023 and early 2024. Their findings are observational. Not experimental. Their sample size was small. They state those limits themselves. We report their findings as what they are. Documented observations from systematic testing. Not a randomised experiment.

Sources

1. Remove, Reduce, Inform: New Steps to Manage Problematic Content, April 2019. About Facebook. Meta's public announcement of the three-part framework. Retrieved 17 August 2026.

1. Reducing the distribution of problematic content, stamped with no specific date. Meta Transparency Center. Source of the current description of reduce as part of the remove/reduce/inform framework and the statement that content may not violate Community Standards but still get reduced distribution. Retrieved 17 August 2026.

1. Types of content that we demote, stamped with no specific date. Meta Transparency Center. Source of the borderline content definition and the four examples quoted above. Retrieved 17 August 2026.

1. Recommendations on Instagram: What creators need to know, 8 June 2023. Instagram for Creators. Source of the two-tiered system description and the distinction between Community Guidelines and Recommendation Guidelines. Retrieved 17 August 2026.

1. Leon Yin and Alfred Ng, Demoted, Deleted, and Denied: There's More Than Just Shadowbanning on Instagram, 25 February 2024. The Markup. The investigation's main findings article. Sample sizes: 70 test accounts, 200+ posts, two custom hashtags. Method: systematic testing with control and experimental conditions. Main finding: war imagery 8.5 times more likely to be hidden from hashtag pages than other content. Authors' stated limitations: brand-new accounts, same IP addresses, small sample size, no statistical significance testing, policies changed during research period. Retrieved 17 August 2026.

1. Leon Yin and Alfred Ng, How We Investigated Shadowbanning on Instagram, 25 February 2024. The Markup. Full methodology including sample sizes, limitations, and experimental design. Retrieved 17 August 2026.

1. Shadow banning: a critical perspective on the limits of visibility in social media recommendation systems, 2024. PubMed Central. Academic research on shadowbanning mechanisms and effects. Retrieved 17 August 2026.

1. Adam Mosseri, Instagram Q&A, 2020. Quoted in Instagram's CEO Says Shadow Banning 'Is Not A Thing.' That's Not True, HuffPost, 25 February 2020. Mosseri's statement about shadowbanning not existing. Retrieved via web search 17 August 2026.

1. Karissa Bell, Even Meta's Oversight Board thinks its rules for banning accounts are baffling, Engadget, 4 June 2026. Reports that Meta "will remove their account from recommendations (which Instagram users often refer to as a 'shadowban')" as an intermediate penalty. Retrieved 21 August 2026.

Changelog

• 2026-08-18: First version drafted.